Can We Talk? A Voice Assistant for Everyone
Voice assistants may be getting smarter but they’re not getting any more interesting to listen to. In fact, their voices all pretty much sound the same. The question is: Why should an in-car assistant sound like a branded voice that everybody else is using?
At Mitsubishi Electric, we’re challenging the status quo of branded voice assistants by offering a new level of customization. With our solution, drivers can personalize the voice of their in-car assistant by using their own voice or another familiar voice – perhaps that of their fiancé, their child, or a close friend – to make their assistant sound more familiar and give them a unique identity that no one else has. Drivers can also use one voice for a while and then choose another, eliminating the dullness that comes from hearing the same voice over and over.
A well-crafted and great-sounding voice is a cornerstone of a great in-vehicle user experience. At the Paris Motor Show, Peugeot announced a voice-activated virtual assistant suitable for autonomous driving in their new “e-Legend” concept car. As its voice, Peugeot adopted a digitized version of Gilles Vidal, its Director of Styling, and enabled “him” to speak in 17 languages.
Now you’re probably wondering about the complexity and the amount of time it takes to create a new voice. In the past, the amount of time was significant – a couple months of work – and involved many hours of speech recordings by a professional voice actor, in a professional studio, under the supervision of a linguistic expert. But this is no longer the case due to new synthetic voice generation algorithms, a direct result of the new machine learning evolution of the last few years.
Our next-gen infotainment solution, FlexConnect.AI uses a deep neural network (DNN) based solution that lets us create new voices on just a few minutes of speech recordings. This enables automakers to differentiate their offering in comparison to solutions from the major assistant providers.
As an example of what a customized voice can sound like, I’ve trained the system on my voice and fed it a simple text sentence to speak (WAV file). Nearly as good as the original!
Please contact us to schedule a private demonstration at TU-Automotive Detroit 2019, where you can see this exciting technology – and more!
